Archived front page, Tue 29 Sep 2026, 07:00 UTC. All of 2026-09-29 · Archive · Live page

The Daily Markdown

The front page of the internet for AI users · Tue 29 Sep 2026, 07:00 UTC

Top now

6 stories

What is climbing fastest across every source

OpenAI6.1xLearning6.8xWorld Labs7.5xGeneral Agents6.9xAnthropic6.9xClaude Sonnet 5.57.0xOpus 5.57.0xAMD7.3xOn-Policy Self-Distillation6.7xNvidia6.2x

Claude Sonnet 5.5

191 points and 138 comments on Hacker News, linking anthropic.com.

Hacker News+ Simon Willison13h ago▲ 9065 to 329 across 4 pollsdiscussanthropic.com

Improving Test-Time Scaling with Adaptive Looped Transformers

Researchers introduced TaH2, a method that directs extra computation to specific tokens during decoding. This approach improves the efficiency and accuracy of test-time scaling on AIME benchmarks. ·

HF Daily Papers7h ago▲ 743 to 30 across 3 pollsarxiv.org

KKKKhazix/AIHOT

A website framework automatically identifies trending topics and generates daily reports. Users can customize its sources and selection criteria to create industry-specific news sites. ·

GitHub7h ago▲ 74462 to 1675 across 6 pollsgithub.com

anthropics/claude-code v2.1.284

Anthropic released Claude Code v2.1.284, making Claude Sonnet 5.5 the default model with a 1M token context window. ·

GitHub Releases13h ago▲ 5830 to 44 across 6 pollsgithub.com

Claude Sonnet 5.5

191 points and 138 comments on Hacker News, linking anthropic.com.

Hacker News+ Simon Willison13h ago▲ 12965 to 329 across 4 pollsdiscussanthropic.com

KKKKhazix/AIHOT

A website framework automatically identifies trending topics and generates daily reports. Users can customize its sources and selection criteria to create industry-specific news sites. ·

GitHub7h ago▲ 90462 to 1675 across 6 pollsgithub.com

Improving Test-Time Scaling with Adaptive Looped Transformers

Researchers introduced TaH2, a method that directs extra computation to specific tokens during decoding. This approach improves the efficiency and accuracy of test-time scaling on AIME benchmarks. ·

HF Daily Papers7h ago▲ 903 to 30 across 3 pollsarxiv.org

anthropics/claude-code v2.1.284

Anthropic released Claude Code v2.1.284, making Claude Sonnet 5.5 the default model with a 1M token context window. ·

GitHub Releases13h ago▲ 8430 to 44 across 6 pollsgithub.com

Trending topics today: heat

OpenAI: 2.77OpenAI2.77Learning: 1.67Learning1.67World Labs: 1.59World Labs1.59General Agents: 1.29General Agents1.29Anthropic: 1.27Anthropic1.27Claude Sonnet 5.5: 1.25Claude Sonnet 5.51.25
Table
OpenAI2.77
Learning1.67
World Labs1.59
General Agents1.29
Anthropic1.27
Claude Sonnet 5.51.25

Heat = summed hotness of today's stories naming the topic.

Models

5 stories

Releases, papers, trending weights and new listings

Top 6 trending on Hugging Face: likes

convaiinnovations/laya: 4,367convaiinnovations/laya4,367Qwen/Qwen-Image-2.1: 2,605Qwen/Qwen-Image-2.12,605abenzerps/Qwen-Image-2.1-Uncensored-GGUF: 2,303abenzerps/Qwen-Image-…2,303XingChen-AGI/Xing4.0-29B-A4B: 1,802XingChen-AGI/Xing4.0-…1,802Edge0/Audio8-ASR-Infinite: 1,452Edge0/Audio8-ASR-Infi…1,452XingChen-AGI/TeleOCR: 826XingChen-AGI/TeleOCR826
Table
convaiinnovations/laya4367
Qwen/Qwen-Image-2.12605
abenzerps/Qwen-Image-2.1-Uncensored-GGUF2303
XingChen-AGI/Xing4.0-29B-A4B1802
Edge0/Audio8-ASR-Infinite1452
XingChen-AGI/TeleOCR826

PirateBench: score out of 100

opus-5-5: 91opus-5-591gpt-6-sol: 82gpt-6-sol82deepseek-v4-flash-0731: 63deepseek-v4-flash-073163
Table
opus-5-591.15
gpt-6-sol81.78
deepseek-v4-flash-073162.82

Agents build our own projects from a brief; every run, failures included: https://cowerx.dev/daily/piratebench/

Improving Test-Time Scaling with Adaptive Looped Transformers

Researchers introduced TaH2, a method that directs extra computation to specific tokens during decoding. This approach improves the efficiency and accuracy of test-time scaling on AIME benchmarks. ·

HF Daily Papers7h ago▲ 33 to 30 across 3 pollsarxiv.org

Claude Sonnet 5.5

191 points and 138 comments on Hacker News, linking anthropic.com.

Hacker News+ Simon Willison13h ago▲ 9065 to 329 across 4 pollsdiscussanthropic.com

Improving Test-Time Scaling with Adaptive Looped Transformers

Researchers introduced TaH2, a method that directs extra computation to specific tokens during decoding. This approach improves the efficiency and accuracy of test-time scaling on AIME benchmarks. ·

HF Daily Papers7h ago▲ 743 to 30 across 3 pollsarxiv.org

DepthBench: Measuring How Residual Connections Enable More Computational Depth

Researchers introduced DepthBench, a benchmark that varies the width-to-depth ratio across ten architectures. They found that deeper models yield benefits only for specific residual connection designs, while standard normalization methods often degrade performance. ·

HF Daily Papers7h ago▲ 733 to 14 across 2 pollsarxiv.org

Claude Sonnet 5.5

191 points and 138 comments on Hacker News, linking anthropic.com.

Hacker News+ Simon Willison13h ago▲ 12965 to 329 across 4 pollsdiscussanthropic.com

Improving Test-Time Scaling with Adaptive Looped Transformers

Researchers introduced TaH2, a method that directs extra computation to specific tokens during decoding. This approach improves the efficiency and accuracy of test-time scaling on AIME benchmarks. ·

HF Daily Papers7h ago▲ 903 to 30 across 3 pollsarxiv.org

DepthBench: Measuring How Residual Connections Enable More Computational Depth

Researchers introduced DepthBench, a benchmark that varies the width-to-depth ratio across ten architectures. They found that deeper models yield benefits only for specific residual connection designs, while standard normalization methods often degrade performance. ·

HF Daily Papers7h ago▲ 893 to 14 across 2 pollsarxiv.org

X voices

5 stories

X posts other sources are linking to, then named voices on Bluesky and Mastodon

GitHub

5 stories

New repos, releases and changelog

New AI repos this week: stars

yetone/magpie: 1,722yetone/magpie1,722KKKKhazix/AIHOT: 1,675KKKKhazix/AIHOT1,675dzhng/jevgrep: 1,522dzhng/jevgrep1,522mikehasa/golive-skill: 1,055mikehasa/golive-skill1,055ollaya-dev/ollaya: 901ollaya-dev/ollaya901feitangyuan/onetake: 885feitangyuan/onetake885
Table
yetone/magpie1722
KKKKhazix/AIHOT1675
dzhng/jevgrep1522
mikehasa/golive-skill1055
ollaya-dev/ollaya901
feitangyuan/onetake885

KKKKhazix/AIHOT

A website framework automatically identifies trending topics and generates daily reports. Users can customize its sources and selection criteria to create industry-specific news sites. ·

A new repository written in TypeScript with 1,675 stars in its first week.

GitHub7h ago▲ 3462 to 1675 across 6 pollsgithub.com

langchain-fireworks==1.7.0

LangChain released version 1.7.0 of its Fireworks integration. The update adds prompt caching middleware and fixes mid-stream read timeout classification. ·

GitHub Releases10h ago▲ 0github.com

anthropics/claude-code v2.1.284

Anthropic released Claude Code v2.1.284, making Claude Sonnet 5.5 the default model with a 1M token context window. ·

GitHub Releases13h ago▲ 030 to 44 across 6 pollsgithub.com

KKKKhazix/AIHOT

A website framework automatically identifies trending topics and generates daily reports. Users can customize its sources and selection criteria to create industry-specific news sites. ·

A new repository written in TypeScript with 1,675 stars in its first week.

GitHub7h ago▲ 74462 to 1675 across 6 pollsgithub.com

anthropics/claude-code v2.1.284

Anthropic released Claude Code v2.1.284, making Claude Sonnet 5.5 the default model with a 1M token context window. ·

GitHub Releases13h ago▲ 5830 to 44 across 6 pollsgithub.com

unslothai/unsloth Laya Decision Models + Library

Unsloth added decision model support, a document library, and Apple Silicon optimizations. The update also enables local Laya execution and accelerates image and video generation by up to 4.5 times. ·

GitHub Releases16h ago▲ 473 to 14 across 8 pollsgithub.com

KKKKhazix/AIHOT

A website framework automatically identifies trending topics and generates daily reports. Users can customize its sources and selection criteria to create industry-specific news sites. ·

A new repository written in TypeScript with 1,675 stars in its first week.

GitHub7h ago▲ 90462 to 1675 across 6 pollsgithub.com

anthropics/claude-code v2.1.284

Anthropic released Claude Code v2.1.284, making Claude Sonnet 5.5 the default model with a 1M token context window. ·

GitHub Releases13h ago▲ 8430 to 44 across 6 pollsgithub.com

unslothai/unsloth Laya Decision Models + Library

Unsloth added decision model support, a document library, and Apple Silicon optimizations. The update also enables local Laya execution and accelerates image and video generation by up to 4.5 times. ·

GitHub Releases16h ago▲ 753 to 14 across 8 pollsgithub.com

Data centers & energy

5 stories

Builds, power deals and the grid

Data-center and energy stories per day

Wed 23: 3Wed 233Thu 24: 3Thu 243Fri 25: 20Fri 2520Sat 26: 1Sat 261Sun 27: 1Sun 271Mon 28: 20Mon 2820Tue 29: 0Tue 290
Table
Wed 233
Thu 243
Fri 2520
Sat 261
Sun 271
Mon 2820
Tue 290

Stories from DCD, Data Center Knowledge, EIA and Hacker News, by publication day.

Can Supercapacitors Smooth AI’s Power Swings?

Supercapacitors offer millisecond response to AI power fluctuations. Experts view them as complements to existing battery and UPS systems rather than replacements. ·

Data Center Knowledge16h ago▲ 0datacenterknowledge.com

Can Supercapacitors Smooth AI’s Power Swings?

Supercapacitors offer millisecond response to AI power fluctuations. Experts view them as complements to existing battery and UPS systems rather than replacements. ·

Data Center Knowledge16h ago▲ 16datacenterknowledge.com

Hardware prices

2 stories

GPUs, memory, and what tokens cost

Output price, $ per million tokens (newest models)

claude-sonnet-5.5: 10claude-sonnet-5.510claude-sonnet-5.5:batch: 5claude-sonnet-5.5:bat…5perceptron-mk1.5: 1.50perceptron-mk1.51.50ember-1: 15ember-115glm-5.3-prime: 8.80glm-5.3-prime8.80median, all paid models: 2median, all paid mode…2
Table
claude-sonnet-5.510
claude-sonnet-5.5:batch5
perceptron-mk1.51.5
ember-115
glm-5.3-prime8.8
median, all paid models2

434 paid models listed; median of the 56 added in 30 days: $4.00. Rankings data by OpenRouter, CC BY 4.0.